Bioequivalence and Bioavailability Forum

Dear Kotu,

welcome to the club! Your observation is common in evaluating partial replicate designs with the FDA guidance code. The source of this is that the model is over-specified. You can't get a reliable estimate for the within-subject variance for the Test formulation in a design in which only the Reference is replicated. See discussions about this in this thread or this one for instance. You may find even more using Search.

Ways out? Don't know exactly . Leaving out the random statement only for the reason that then it gives a result is IMHO not a reasonable solution.
You may try to simplify the between-subject variance-covariance part of the model from FA0(2) to CS i.e. the RANDOM statement to
RANDOM TRT/TYPE=CS SUB=SUBJ G;
Sometimes this helps. But it bears the assumption that the between-subject variances of Test and Reference are the same and neglects the subject-by-formulation interaction, which is different from the original model.

Another model specification with SAS code may be found in this thread.

But as already said: Don't know exactly the correct way out.
Eventually you may consider to ask the FDA statisticians. Barbara Davit is a pleasant-natured woman as Helmut told me once.